The jury returned a unanimous guilty verdict.
Guilty of being a remarkable piece of hardware, but also guilty of being a hard-nut-to-crack platform.
Kudos to the team
But seriously, who am I to judge? With every new post in this series, I feel less and less confident to make any more statements about Epiphany or Paralella board. Why?
As I go deeper and deeper into the technical stuff, I unravel the true/hidden story of the company. True, in the sense that I see more of the human or business part of the endeavor, and that part is getting more impressive with every slide deck, blog post, or presentation I happen to find on the internet.
Of course, I could focus only on the final effect; this is the state of the compiler, libraries, or operating system integration and rant about things that could’ve been done better. Yes, yes … we all know it’s all crucial for the business, but not seeing the bigger picture would lead us to very unfair conclusions!
So what’s the non-technical backstory? – The handful of people building a hardware platform (not only chip!) from scratch, successfully releasing it to the world while taking substantial financial risks and working hard around the clock.
Have you ever worked with other non-mainstream SBCs? Likely not, but most for sure you have a Raspberry PI or OpenWRT router plugged in at home. If you’re not in academia or the hardware industry, you won’t have an opportunity to test out many new architectures, etc.
Therefore I would like to raise a cup, and say kudos to the Adapteva Team for the fantastic work. Without the Kickstarter campaign, I wouldn’t have a chance to write this miniseries and learn so much about the hardware.
From the bottom of my heart, I admire Andreas Olofsson as an Engineer, and I wish to work with people as passionate as him in the future.
Sell solutions, not components (source)
I think that’s the ultimate lesson we learned here. Epiphany chip can do a lot of things, but what problem does it solve?
When I bought parallella, there have been a few examples available in the repo so that I could use them out of the box. However, the demo applications were just nice demos, nothing that I could compile and deploy to production. Imagine what you could do if you had an OpenCV example with an object-detection algorithm there? Or some tiny machine learning library available too?
That would win the crowd!
Hardware without software is useless (source)
Who wants to spend time on troubleshooting compiler, kernel, or hardware (Verilog/FPGA) issues? Not many people, definitely not the average Raspberry PI user who might have bought the Paralella board.
Also, code reuse is the key to success. In the Epiphany case, too often, we have to adjust or even rewrite OSS libraries.
Arnd Bergmann shared an excellent presentation on the SoC lifecycle and spoke about the CPU market vision for the upcoming years.
That made me think about how hard and expensive it is to build a new CPU architecture that lasts for years, if not decades. It’s a challenging domain and a competitive business.
Our journey through the stack ends here. I hope you enjoyed the blog posts as much as I did enjoy writing them.
On the personal note, I regret that Parallella never made it to the mainstream, and I hoped it would be the first successful platform to evangelize the many-core architecture in the world. It didn’t happen, so I guess it’s time to move on.
So, if you have heard about other cool, non-mainstream, affordable, and publicly available hardware please let me know in the comments.
See other posts!
- # Parallella (part 1): Case study
- # Parallella (part 10): Power efficiency
- # Parallella (part 11): malloc
- # Parallella (part 12): Tensorflow?
- # Parallella (part 13): Closing notes
- # Parallella (part 2): Hardware
- # Parallella (part 3): Kernel
- # Parallella (part 4): ISA
- # Parallella (part 5): elibs
- # Parallella (part 6): FreeRTOS